We cannot overlook or ignore the effect of science in our life. We use the Internet in many parts of our day-to-day life; for example, to check the weather, to drive to a location, to express our thoughts, and so on. If we try to understand the effect of science in our life precisely, then we will notice that these are the outcome of using Artificial Intelligence and Machine Learning applications.
The terms AI and ML come up very frequently when discussing analytics and the technological changes which are sweeping through our world. Artificial Intelligence is the broader concept of machines being able to carry out tasks in a way that we would consider “smart”, and Machine Learning is an application of AI, based around the idea that we should simply be able to give machines access to data and let them learn for themselves.
This article is based on the Tech 2030 podcast episode “The EverydAI Intelligence.” Click here to listen
Dr. Detlef Nauck, Head of AI & Data Science Research for British Telecom’s Applied Research Division, explains why these systems are used. He leads a programme of about 30 international researchers who develop capabilities underpinning modern AI systems.
“You can use these systems for two reasons. One, you want to make decisions cheaper and faster and take humans out of the loop. Two, you want to augment the human capability to make decisions, typically in areas like medicine. These things are called human-in-the-loop systems, also intelligent augmentation. The idea is you’d use AI to Improve the human capabilities in decision-making,” Nauck explains.
As technology and our understanding of how our minds work has progressed, our concept of what constitutes AI has changed. Rather than increasingly complex calculations, work in the field of AI concentrates on mimicking human decision making processes and carrying out tasks in ever more human ways.
Let’s say you’re making a self-driving car and want it to stop at stop signs. You would need Data Science, AI, and ML to make it possible.
Machine learning is used to make the car recognise stop signs using cameras. You’ll need to create a dataset with street side object pictures and train an algorithm to recognise those with stop signs on them.
Artificial intelligence is when the car recognises the sign. The vehicle should hit the brakes right on time, not too early and not too late.
Let’s imagine that, while running a test, we see that the car doesn’t react to stop signs sometimes. What do we do?
Maybe the car misses stopping signs at night because the training data only has objects in daylight. In this case, we add a few night-time photos and get back to testing.
That’s how the whole ML vs. AI. vs. data science correlation works. As you see, they all go hand in hand: machines won’t learn without data, and it’s better to do data science with machine learning.
Let’s Delve into the AI and ML Use Cases
Artificial Intelligence, and in particular ML, certainly has a lot to offer. With its promise of automating mundane tasks, as well as offering creative insight, industries in every sector – from banking to healthcare and manufacturing – are reaping the benefits.
“The benefits of using AI, machine learning, data science depends on the area. Typically, you want to make decisions based on data,” Nauck says. “Let’s say you want to make a decision about a loan application, and you want to do this automatically. That means you have data about the application. Should the loan be granted? Yes? No? In order to do this, you need to create a machine learning model from the status for learning process, and that would become your artificial intelligence solution.
“You can also think of a context where this decision-making has to be very fast. In a network context, you use these decisions to route traffic, and obviously you have an artificial intelligence system that has to make decisions about traffic routing in maybe a millisecond kind of reaction time.”
Machine learning is pushing data science to the next level of automation. AI is about human-AI interaction gadgets like Siri, Alexa, Google Home, and many others. The video and audio prediction systems like those of Netflix, Amazon, Spotify, and YouTube are ML-powered.
The development of neural networks has been key to teaching computers to think and understand the world in the way we do, while retaining the innate advantages they hold over humans such as speed and accuracy.
A Neural Network is a computer system designed to work by classifying information in the same way a human brain does. For example, it can be taught to recognize images and classify them according to elements they contain.
However, the training data provided to the computers for Machine Learning has a great impact on the accuracy of the system. If you feed the system with biased content…
“Bias means that a group in your data would be underrepresented or maybe overrepresented. We know that from facial image recognition the AI systems that have been developed [in many countries] are heavily biased in the sense that they are recognizing images of people with dark skin or females worse than the images of white males. This racial and gender bias has crept into these systems because the data that was used to train them did not represent the population at the right level.
The problem with bias is that you need to look very, very carefully at the data to train a machine learning model. A machine learning model will always pick up on the bias in the data because, essentially, this is what these learning algorithms do. They try to exploit information that helps them make better decisions.”
Machine learning methods and tools are used extensively in the medical field – to detect a disease, therapy planning, prediction of its progress etc. Other applications include image recognition, speech recognition, autonomous driving, virtual personal assistant, recommendation of products and customer service.
“In the telecom industry, you have a number of application scenarios to improve interactions with customers; by improving workflow management systems, using AI to make automated decisions within operational workflows,” Detlef Nauck explained.
“Using AI to predict faults in networks, [to perform] predictive maintenance, rerouting traffic or switching off certain parts of the network. These things can be done easily. And you can very readily apply machine learning for different scenarios. You can use it to detect anomalies in your network, or cyber-attacks and fight them off.
“We have a product [at BT] that tries to stop nuisance callers reaching you. We profile the calling traffic that comes into the network and try to detect behaviour patterns that indicate a spammer or nuisance calls,” he said.
Challenges for the Future
Ongoing research activities are taking into consideration diverse aspects, such as the availability and usability of datasets needed for specification and testing of AI/ML solutions, regulatory aspects, and practical implementation issues. Telecommunications networks generate a huge amount of data. Is this data useful and applicable?
“The challenge in that space is that you typically don’t have data that is labelled in a way that tells you ‘This is what you should do’. You will find scenarios where the network behaves in a certain way,” Nauck explained.
“You may get into scenarios where you need to look at the more classical techniques like optimisation and search that help you to identify configurations in the network that are beneficial. You have AI programs that optimise local areas of the network, but they need to collaborate to optimise the overall behaviour of the network. And these kinds of techniques are still in development.”
There are many challenges around automating a network via enabling components to directly contribute into decision-making activities related with the mobile network resource management.
“Controlling a network or automating a network with AI is a big challenge and I think that is largely unaddressed at the moment. We need to look into the methods that have been used in AI for decades for around optimisation, need to think how we do local optimisation and collaborative scenarios of intelligent agents to influence the overall network and go to something that is more like goal-oriented or intent driven. The idea that we can just take all the data and run it through a machine learning algorithm is not going to work,” the expert believes.
Despite the hype of the previous few years, the adoption of AI/ML methods in cellular networks is still at its early stages. A lot of work is still needed to identify the most suitable solutions for the dynamic network management and control via AI/ML mechanisms.
“I think it’s important to realise that there are AI areas overhyped still, and this comes from the success that we see in natural language processing or image analysis and the applications of deep networks in these spaces. And we hear news about these systems growing and growing and getting up to billions of parameters that need to be tuned, which will cost you millions of dollars to do it in cloud environments. These things cannot be scaled indefinitely,” Nauck said.
“In any AI topic, it helps to have highly diverse teams. People from all over disciplines, but also from different backgrounds to get the ideas flowing and circulating and to not forget certain inputs.”
A key part of Dr Nauck’s work is to establish best practices in Data Science and Machine Learning for conducting data analytics professionally and responsibly. He has a keen interest in AI Ethics and Explainable AI to tackle bias and to increase transparency and accountability in AI. What does he think the future has in store for this field?
“I mean, it’s always difficult to make predictions, especially when they’re about the future. What I would imagine is we will see steady progress in areas like natural language processing. I would expect that in 10 years’ time, we’ll have types of personal translators, so that talking to somebody else in a different language won’t be an issue any more because we’ll have enough language data that we could use to train up translators.
“In terms of autonomous devices, I think we will be seeing much more of them. If you look into the IoT space, then the development of smart devices will definitely continue. We’ll see more discussions around AI ethics and what can be done and should be done with AI, especially when it goes into devices that are around you and your house. The typical stuff like “will we see autonomous cars in the street or robots that do our dishes?” is a bit overhyped,” he concluded.
Written by Renuka Racha